Flood Detection in ENVI

Project Objective:

The objective of this project was to analyze where flooding from Hurricane Ian occurred in the Orlando metropolitan area. 

Data Collection:

Both the pre-flood and post-flood satellite imagery collected was from the Sentinel-2 satellite, found online via the Copernicus Browser website.  The imagery was atmospherically corrected using L2A processing and was projected into WGS 1984 UTM Zone 17N. The AOI is approximately 35.46km by 21.12km big, covering the downtown and most of the surrounding suburbs.

Resolution:

• Spatial Resolution: 14m

• Temporal Resolution:  5 days

• Spectral Resolution: 13 bands

• Radiometric Resolution: 12-bit

AOI Coordinates (Decimal Degrees):

• NW Corner:  -81.332124584 W,  28.384655219 N 

• NE Corner:   -81.190683 W,   28.642087 N

• SW Corner:   -81.552544 W,   28.45043 N

• SE Corner:   -81.19034 W,   28.451419 N

Methodology:

1. Downloaded the Sentinel-2 imagery from the Copernicus Browser website. The imagery came in the form of several raw bands.

2. Uploaded the imagery into ArcGIS Pro.  Used the "Composite Bands" tool in ArcGIS Pro to combine every raw band data file into one raster file. Did this process to create one TIFF raster file for the July imagery and the September imagery. Performing this step allowed me to set any band to display as red, green, or blue.  

2a.  Since my imagery was downloaded in the size of my AOI, subsetting and mosaicking was not needed, although those steps would have been performed in ArcGIS Pro if I needed to. The imagery was also pre-projected into WGS 1984 UTM Zone 17N, so projecting wasn't needed as well for this project. 

3. Imported the composited TIFF files into ENVI. Used the "Change Detection Workflow" tool to carry out my analysis.

3a. Chose my July pre-flood imagery as "Input Raster 1" and my September post-flood imagery as "Input Raster 2".

3b. When prompted to either perform or skip Image Registration, I chose to perform it. 

3c. Set the "Method" to "Band Difference". When asked to select my input band, I chose "Band 9", which corresponded to Sentinel-2's B8A Narrow NIR sensor. 

3d. During the "Threshold" and "Vectorize Changes" steps, I left the default options for each step.

3e. On the "Export Results" step, I opted to export the change detection raster as both a Classification Raster and a Shapefile

ArcGIS Pro and ENVI Tools Used in Analysis:

ArcGIS Pro:

• Composite Bands

• If Necessary:

  • Project Raster

  • Mosaic to New Raster

  • Clip Raster

ENVI:

• Change Detection Workflow

Imagery:

Results:

The biggest pockets of flooding occurred near the northeast and southeast corners, as well as the center-left portion of the study area. Most of the flooding was observed in the suburban areas of Orlando. In downtown Orlando, most of the flooding happened near the major downtown lakes.

While Orlando generally has good flood prevention infrastructure in place, the flood detection from Hurricane Ian shows that the city (and Orange County at-large) can perform better. Future investment into flood prevention measures should focus on the areas in red. Targeted investment will lower the amount of flooding that occurs the next time a Hurricane brings a historic amount of rainfall to the region. With less flooding, more vehicles can access more areas within a shorter period of time, enabling first-responders to rescue people from flooded areas and ambulances to transport the injured to hospitals. That will save lives in the long run as Central Florida experiences more and stronger hurricanes every year due to climate change.